I went on a long journey trying to find a pebble replacement. If you are looking for the closet smart watch to the pebble experience then https://a.co/d/6hEwdmc check out the Garmin instinct 2. 28 day battery life. Gets notifications from phone. Eink like screen. Its my daily wear
Post by @naval on January 3:
"AI won't replace programmers, but rather make it easier for programmers to replace everyone else."
Reply by @X.Wallets on January 3:
"AI can write programs per CEO Jensen Huang."
Follow-up post by @naval on January 3:
"Calculators can do math per CEO of Texas Instruments."
---
This exchange has been bouncing around my head ever since it happened. CEOs especially those who are investing heavily in AI keep saying stuff that does not align with you my experiences working with and building these AI systems to the point where I think that people are saying things to justify their own investment in technology.
This is a CEO selling themselves and their company nothing more.
You should be able to do that already by taking all of your customers nit embeddings and averaging them to produce a point in space that represents the universal nit. Embeddings are really cool and the fact that they still work when averaging is one of their cool properties.
This isn't my post but I work with llms everyday and I've seen this kind of instruction ignoring behavior on sonnet when the context window starts getting close to the edge.
I don't know, in practice there are so many potential causes that you have to look case by case in situations like that. I don't have a ton of experience with the raw Claude model specifically, but would anticipate you'll have the same problem classes.
Usually it comes down to one of the following:
- ambiguity and semantics (I once had a significant behavior difference between "suggest" and "recommend", i.e. a model can suggest without recommending.)
- conflicting instructions
- data/instruction bleeding (delimiters help, but if the span is too long it can loose track of what is data and what is instructions.)
- action bias (If the task is to find code comments for example, even if you tell it not to, it will have a bias to do it as you defined the task that way.)
- exceeding attention capacity (having to pay attention to too much or having too many instructions. This is where structures output or chain of thought type approaches help. They help focus attention on each step of the process and the related rules.)
I feel like these are the ones you encounter the most.
Yes and no. I've found the order in which you give instructions matters for some models as well. With LLMs, you really need to treat them like black boxes and you cannot assume one prompt will work for all. It is honestly, in my experience, a lot of trial and error.
Learning from copy written work requires a license to access that work. You can extract information from the world's best books by purchasing those books. But no author is being compensated here they download books.torrent then uses that pirated material to then profit.
I have to wonder if that design came about because Apple didn’t want people to leave the mouse plugged in all the time. It’s the kind of reasoned institutional blunder that I feel like I encounter in my work life fairly often out of momentum and groupthink.
I suspect the conversation was closer to “make it rechargeable, but don’t change anything else.”
You obviously can make an Apple style mouse that plugged in sensibly, but that would require changing the case and internal layout. The way they did it the new battery and charge port fit in the space left by the old AA’s.
If you want to further add vertical resolution, Unicode 13 introduces 2x3 mosaics (from the TRS-80, Teletext and others) that are supported directly within VTE (so people don't need to install a font). Slightly more complicated are the "smoothed" mosaics (from Minitel 2, IIRC) but they offer diagonal lines (and complement PETSCII and ATASCII symbols also introduced in version 13).
I expect 2x4's to be available for version 14, but it'll be a while.
2x3's (and 2x4's) are kind of funny, because you still won't be able to use more than 2 colors per cell, so it works for cells with two colors but that can be subdivided in thirds (or quarters). Finding a best fit for a cell and a mosaic is not a trivial job.
It's a shame support is relatively limited. Apple's Terminal doesn't (even though it passes the VT-100 torture test, which is, in itself, remarkable), nor does the VTE-based ones common in other unixes.
